View
0
Download
0
Category
Preview:
Citation preview
RSCAS 2019/30 Robert Schuman Centre for Advanced Studies Migration Policy Centre
Estimating Transnational Human Mobility on a Global
Scale
Ettore Recchi, Emanuel Deutschmann and Michele Vespe
European University Institute
Robert Schuman Centre for Advanced Studies
Migration Policy Centre
Estimating Transnational Human Mobility on a Global Scale
Ettore Recchi, Emanuel Deutschmann and Michele Vespe
EUI Working Paper RSCAS 2019/30
This text may be downloaded only for personal research purposes. Additional reproduction for other
purposes, whether in hard copies or electronically, requires the consent of the author(s), editor(s).
If cited or quoted, reference should be made to the full name of the author(s), editor(s), the title, the
working paper, or other series, the year and the publisher.
ISSN 1028-3625
© Ettore Recchi, Emanuel Deutschmann and Michele Vespe, 2019
Printed in Italy, April 2019
European University Institute
Badia Fiesolana
I – 50014 San Domenico di Fiesole (FI) Italy
www.eui.eu/RSCAS/Publications/
www.eui.eu
cadmus.eui.eu
Robert Schuman Centre for Advanced Studies
The Robert Schuman Centre for Advanced Studies, created in 1992 and currently directed by
Professor Brigid Laffan, aims to develop inter-disciplinary and comparative research on the
major issues facing the process of European integration, European societies and Europe’s place
in 21st century global politics.
The Centre is home to a large post-doctoral programme and hosts major research programmes,
projects and data sets, in addition to a range of working groups and ad hoc initiatives. The
research agenda is organised around a set of core themes and is continuously evolving,
reflecting the changing agenda of European integration, the expanding membership of the
European Union, developments in Europe’s neighbourhood and the wider world.
For more information: http://eui.eu/rscas
The EUI and the RSCAS are not responsible for the opinion expressed by the author(s).
Migration Policy Centre (MPC)
The Migration Policy Centre (MPC) is part of the Robert Schuman Centre for Advanced Studies at the
European University Institute in Florence. It conducts advanced research on the transnational
governance of international migration, asylum and mobility. It provides new ideas, rigorous evidence,
and critical thinking to inform major European and global policy debates.
The MPC aims to bridge academic research, public debates, and policy-making. It proactively engages
with users of migration research to foster policy dialogues between researches, policy-makers, migrants,
and a wide range of civil society organisations in Europe and globally. The MPC seeks to contribute to
major debates about migration policy and governance while building links with other key global
challenges and changes.
The MPC working paper series, published since April 2013, aims at disseminating high-quality research
pertaining to migration and related issues. All EUI members are welcome to submit their work to the
series. For further queries, please contact the Migration Policy Centre Secretariat at migration@eui.eu
More information can be found on: http://www.migrationpolicycentre.eu/
Disclaimer: The EUI, RSCAS and MPC are not responsible for the opinion expressed by the author(s).
Furthermore, the views expressed in this publication cannot in any circumstances be regarded as the
official position of the European Union.
Abstract
We devise an integrated estimate of country-to-country cross-border human mobility on the basis of
global statistics on tourism and air passenger traffic. The joint use of these two sources allows us to (a)
test for their relative contribution, and (b) correct for their limitations to the estimate of global mobility
by combining them. The two sources are adjusted and merged following simple procedures. The
resulting dataset, which covers more than 15 billion estimated trips over the years 2011 to 2016,
promises to be a systematic and comprehensive resource on transnational human mobility worldwide.
In this paper, we illustrate the data characteristics and transformations adopted in creating this dataset.
First applications are explored, and its remaining limits are discussed.
Keywords
Transnational human mobility, travel, globalization, network data, tourism, air traffic.
1
1. Introduction
The increase in the cross-border mobility of persons—with differing reasons and objectives—is a
hallmark of the current age of human history. In the face of this spectacular social trend, which is in
place since at least the 1950s, there is a surprising dearth of systematic information detailing the size of
travel flows across countries. The Global Mobilities Project (GMP) at the European University
Institute’s Migration Policy Centre (MPC) intends to fill this gap by addressing different dimensions of
transnational mobilities (Recchi 2017).1 As regards the volume and directions of mobility flows, we
capitalize on two of the most comprehensive sources of transnational human movements at the global
scale:
1. Data on tourism, i.e., cross-border visits that include an overnight stay (nota bene: not necessarily
for leisure), from the World Tourism Organization (UNWTO);
2. Data on cross-border air passenger traffic from Sabre, a travel industry company.
Being conceived and collected for different purposes, both sources, taken individually, have clear
limitations when used in the attempt to provide insights into human global mobility. The data on tourism
is incomplete in that people moving between countries for reasons other than tourism (in particular,
returning residents) are not included. It is also sometimes distorted because visitors from countries with
few departures are not counted since their travel origin does not show up in the receiver country’s
tourism statistics. The data on air passenger traffic, in turn, does not factor in people who do not travel
by airplane. In particular, journeys between neighboring countries, where cross-border mobility is
particularly high (Deutschmann 2016), are likely to be severely underestimated since people often use
car, railway, or bus transportation rather than flights. We propose to remedy these systematic biases by
combining the two data sources, producing more reliable estimates of cross-country human mobility
globally.
In this paper, we first make general considerations about the composition of transnational mobility
flows in the two sources and give an overview of the procedures followed to combine them (section 2).
We describe these procedures in more detail in section 3. Section 4 highlights some findings derived
from first explorations of the newly created dataset. In the conclusion (section 5), we outline some
pending limitations and advocate the use of the novel dataset to study transnational human mobility
empirically in social science research.
2. Understanding the composition of transnational mobility flows
Our aim is to have robust estimates of the absolute number of yearly travels from and to every country
worldwide.2 In formal terms, we set out to measure the volume of cross-border travels T across all pairs
of sovereign states a, b, c, … n on the planet. Such travels are carried out by both non-residents (NR)
and residents (R) of receiving countries and take place by air (flights) or by land/water transportation
1 We use the term ‘transnational’ in the meaning it has in the field of international relations, where it is employed to describe
any movement by non-state actors that spans across national borders (Nye and Keohane 1971). We are aware that in the
field of migration studies ‘transnational’ has a more demanding meaning that involves the regular movement of the same
individuals across certain borders (Wimmer and Glick Schiller 2002). Following the latter tradition, it would be equally
justified to speak about international mobility.
2 Conceptually, migrants and asylum-seekers are excluded from our estimates, even though we cannot rule out that some
‘visitors’ may overstay their travels and thus become migrants and asylum-seekers. More on the issue in the Conclusions
(section 5).
Ettore Recchi, Emanuel Deutschmann and Michele Vespe
2 Robert Schuman Centre for Advanced Studies Working Papers
(trains, buses, cars and private road vehicles, boats, ferries and ships),3 that we indicate respectively with
exponents A and L. Therefore:
𝑇𝑎→𝑏 = 𝑁𝑅𝑎→𝑏𝐴 + 𝑅𝑎→𝑏
𝐴 + 𝑁𝑅𝑎→𝑏𝐿 + 𝑅𝑎→𝑏
𝐿
Unfortunately, no existing source contains information on all four factors simultaneously. The original
tourist files include only 𝑁𝑅𝑎→𝑏𝐴 + 𝑁𝑅𝑎→𝑏
𝐿 , i.e., they register tourist arrivals in destination countries,
but not tourists returning to their countries of origin. Air traffic statistics include 𝑁𝑅𝑎→𝑏𝐴 + 𝑅𝑎→𝑏
𝐴 , i.e.,
air passengers only.4 Thus, both datasets are suboptimal as they systematically exclude 𝑅𝑎→𝑏𝐿 . We expect
the two datasets to be strongly correlated, because they share the same core component: 𝑁𝑅𝑎→𝑏𝐴 . They
should diverge only when 𝑅𝑎→𝑏𝐴 and/or 𝑁𝑅𝑎→𝑏
𝐿 are large and/or not correlated.
The original UNWTO tourist files, however, also record residents of b going from b to a with all
transportation means, that is 𝑅𝑏→𝑎𝐴 and 𝑅𝑏→𝑎
𝐿 . If we imagine that these people return to their country of
residence in the same year of their outbound travel, we can count them as part of 𝑅𝑎→𝑏𝐴
and 𝑅𝑎→𝑏𝐿 . We
can thus assume that 𝑅𝑎→𝑏𝐴 + 𝑅𝑎→𝑏
𝐿 = 𝑅𝑏→𝑎𝐴 + 𝑅𝑏→𝑎
𝐿 . This assumption falls short of a small proportion
of travellers who: a) travel by the end of the year and come back in the following calendar year, or b)
resettle abroad. As for a), given the overall constancy of travel flows, we can maintain that these
travellers are offset by similar travellers twelve months earlier. As for b), these travellers are migrants.
A comparison of migration flows (as estimated by Abel and Sander 2014) and global tourist flows (based
on Deutschmann 2016) shows a 1/150 relationship. That is, migrant travels correspond to about 0.6
percent of tourist travels, which is therefore the approximate overall size of error we introduce in our
tourism estimates with this assumption.5 We therefore revise the original UNWTO tourism data to build
a yearly matrix of tourists/visitors travelling from a to b which also includes (returning) travellers from
b who moved to a:
𝑇𝑎→𝑏𝑟𝑒𝑣𝑖𝑠𝑒𝑑 = 𝑁𝑅𝑎→𝑏
𝐴 + 𝑁𝑅𝑎→𝑏𝐿 + 𝑅𝑏→𝑎
𝐴 + 𝑅𝑏→𝑎𝐿
Hereafter, we will call this the GMP-revised tourism data [1]. Its creation is described in detail in section
3.1.
As for the air passenger data, which we use in its KCMD-revised form [2] (see explanation below),
we assume that they tend to be lower than the revised tourism data [1] because travellers also move with
other transportation means. However, [1] and [2] should converge the larger the distance between origin
and destination as air travel tends to become the exclusive means of transportation at long distances.
This distance-mediated relationship between [1] and [2] leads us to transform the air passenger data. We
compute an estimate of transnational mobility [3] that adjusts [2] by a factor that accounts for the
distance between countries. The formal procedure to estimate [3] is described in section 3.3.
In a final step, we combine the two revised sources, [1] and [3], to create an integrated dataset on
global transnational mobility. As we hold that both [1] and [3] tend to underestimate actual mobility
flows, our final estimate is always the largest of the two when we have both information—that is, either
[1] or [3]. When we lack [3], we take [1], and vice versa.
3 Other statistically marginal forms of mobility (by foot or bike, for instance) are also included, provided they take place
legally (i.e., they are registered). Unregistered or illegal border crossings are in fact left out by default from tourism and air
traffic statistics, and, as a consequence, from our estimates.
4 Note that air traffic statistics do not allow us to distinguish between these two components since they are based on the
location of the airport of origin and destination, not on the residence or nationality of the traveller.
5 As our final estimate of global transnational mobility provides higher figures, the actual migrant/travel ratio is likely to be
even lower, with migrant flows corresponding to about 0.001 percent of travel flows (see section 4, Figure 6).
Estimating Transnational Human Mobility on a Global Scale
European University Institute 3
Figure 1 provides an overview of this procedure. The individual steps are described in more detail in
the following sections. The resulting final dataset covers 196 sender and receiver countries, generating
a symmetric matrix of 38,220 cases (i.e., country pairs) per year. For the entire 2011-2016 period, about
9.5 billion trips (approx. 61 percent) are ultimately derived from [1] and 6 billion trips (approx. 38
percent) from [3]. Overall, 12.0 percent of cells are empty, which can mean either a total absence of
transnational mobility between these countries (most likely in the case of pairs of small and distant
nations) or missing data. The Global Transnational Mobility Dataset covers an estimated total of 15.7
billion trips.
Figure 1. Overview of the data composition
KCMD-revised air
passenger trend data [2]
Only [1]
is available
[1] and [2]
are available
Only [3] is
available
[1] > [3] → [1]
[3] > [1] → [3]
GMP-revised
tourism data [1]
Original tourist files Original air passenger files
Global Transnational Mobility Dataset
Distance-adjusted air
passenger data [3]
[1] = [3]
Ettore Recchi, Emanuel Deutschmann and Michele Vespe
4 Robert Schuman Centre for Advanced Studies Working Papers
3. Constructing the dataset
In the following subsections, we outline in more detail how we handled the raw data and proceeded until
the creation of the final Global Transnational Mobility Dataset. We first describe the creation of the
GMP-revised tourism data (section 3.1). Second, we bring the KCMD-revised air passenger trend data
in (section 3.2). Third, we introduce the correction factor that adjusts the latter source, taking geographic
distance into account (section 3.3). Finally, we describe the merging and finalization of the dataset
(section 3.4).
3.1 Creating the GMP-revised tourism data [1]
Our first and primary source, the UNWTO tourism data, was obtained by the Global Mobilities Project
(GMP) of the EUI’s Migration Policy Centre (MPC) from the UNWTO as a set of files containing yearly
flows from 1995 to 2016 for all sovereign countries and dependent territories worldwide (UNWTO
2015).6 The original data contains 219 excel files, one per receiver country/territory. To create a unified,
standardized, and usable dataset (hereafter the GMP-revised tourism data), we took the following steps:
Step 1: Prioritizing the different UNWTO operationalizations of ‘arrivals’
The country-to-country flow data on arrivals is reported in eight different categories in the UNWTO
data (Table 1). The UNWTO defines arrivals—and describes its sources—as follows:
Arrivals data measure the flows of international visitors to the country of reference: each arrival
corresponds to one inbound tourism trip. If a person visits several countries during the course of a
single trip, his/her arrival in each country is recorded separately. In an accounting period, arrivals
are not necessarily equal to the number of persons travelling (when a person visits the same country
several times a year, each trip by the same person is counted as a separate arrival).
Arrivals data should correspond to inbound visitors by including both tourists and same-day non-
resident visitors. All other types of travelers (such as border, seasonal and other short-term workers,
long-term students and others) should be excluded, as they do not qualify as visitors. Data are
obtained from different sources: administrative records (immigration, traffic counts, and other
possible types of controls), border surveys or a mix of them. If data are obtained from
accommodation surveys, the number of guests is used as estimate of arrival figures; consequently,
in this case, breakdowns by regions, main purpose of the trip, modes of transport used or forms of
organization of the trip are based on complementary visitor surveys. (UNWTO 2015, p. 9)
To include as many cases as possible in the unified dataset, we use all eight ‘arrivals’ categories shown
in Table 1, in order of preference. This preference order is justified on the basis of a number of
assumptions and compromises that are discussed in the Appendix.
Step 2: Creating a unified excel file
We then created a unified excel file that contains the relevant country-to-country flow data to all
countries for which this information was available.7 In doing so, we exclude several ‘odd’ sender
categories, such as ‘Other countries of the world’, which cannot readily be included in a country-to-
country flow matrix. Details about this procedure and its consequences are described in the Appendix.
6 At UNWTO, we thank Jacinta Mora for facilitating our access to these tourism statistics.
7 There are 18 countries that are part of the UNWTO data collection that do not report country-to-country flow data. This
means they may be part of the full tourism dataset as senders of tourists but not as receivers. They are: Afghanistan, Bonaire,
Djibouti, Equatorial Guinea, Eritrea, Gabon, Ghana, Guinea-Bissau, Liberia, Libya, Mauritania, Saba, Sao Tome and
Principe, Sint Eustatius, South Sudan, Syrian Arabic Republic, Turkmenistan, and United Arab Emirates.
Estimating Transnational Human Mobility on a Global Scale
European University Institute 5
Table 1. Categories in the UNWTO dataset
Code Description Preference
112 Arrivals of non-resident tourists at national borders, by country of residence 1st
111 Arrivals of non-resident tourists at national borders, by nationality 2nd
122 Arrivals of non-resident visitors at national borders, by country of residence 3rd
121 Arrivals of non-resident visitors at national borders, by nationality 4th
1912 Arrivals of non-resident tourists in all types of accommodation establishments, by
country of residence
5th
1911 Arrivals of non-resident tourists in all types of accommodation establishments, by
nationality
6th
712 Arrivals of non-resident tourists in hotels and similar establishments, by country of
residence
7th
711 Arrivals of non-resident tourists in hotels and similar establishments, by nationality 8th
Step 3: Adding returning residents
In line with the considerations made in section 2, we add the returning residents 𝑅𝑏→𝑎𝐴 + 𝑅𝑏→𝑎
𝐿 , to the
incoming non-residents 𝑁𝑅𝑎→𝑏𝐴 + 𝑁𝑅𝑎→𝑏
𝐿 to obtain a more complete picture of human mobility across
borders. In doing so, we effectively double the number of trips in the tourism dataset. Furthermore, the
matrix becomes symmetric, i.e., mobility flows are now, by necessity, the same in both directions
(𝑇𝑎→𝑏𝑟𝑒𝑣𝑖𝑠𝑒𝑑 = 𝑇𝑏→𝑎
𝑟𝑒𝑣𝑖𝑠𝑒𝑑). After this step, we have obtained the GMP-revised tourism data [1].
3.2 Bringing the KCMD-revised air passenger trend data [2] in
The second source is the dataset on global air passenger traffic in the 2011–2016 period collected by a
private travel industry company, Sabre (2014). The dataset contains information on the number of air
passengers per month, traveling between airports. Here, we draw on a simplified and reduced version
created by researchers at the European Commission’s Knowledge Centre for Migration and Democracy
(KCMD) that represents the yearly trend between countries (henceforth KCMD-revised air passenger
trend data [2]). This version was generated through a time-series decomposition that dissects the raw
overall air passenger flow between two countries into a trend component, a seasonal component, and a
residual component (Gabrielli et al. 2019). In the KCMD-revised air passenger trend data [2] used here,
the monthly trend data is aggregated to yearly averages. The data is available for the years 2011 to 2016.
We merge the two datasets [1] and [2] using ISO 3166-1 alpha-3 country codes. In line with the
considerations made in section 2, we hypothesize:
a) [1] to be on average larger than [2], as it includes both air passengers and land/water
travellers;
b) [1] and [2] to be highly correlated, since many travellers use flights to cross borders;
c) [1] and [2] to be more strongly correlated as the distance between country pairs increases,
since people are more likely to use air transportation at longer distances.
All three hypotheses hold empirically. As expected, tourism figures based on [1], reporting cross-border
trips with all transportation means, tend to be higher than air passenger figures based on [2], reporting
journeys that take place with flight transportation only. The exceptions are mainly countries receiving
by plane a number of returning residents or nationals exceeding the number of non-national visitors (de
facto, out-migration countries with little incoming tourism). Table 2 shows the distribution of the
deviations between the two data sources across cases (i.e., country pairs), by year. Negative values
denote that there are more tourists than air passengers; positive values denote that there are more air
passengers than tourists travelling between a pair of countries. The average median (50th percentile)
Ettore Recchi, Emanuel Deutschmann and Michele Vespe
6 Robert Schuman Centre for Advanced Studies Working Papers
across years is -2,410 trips, and even at the 75th percentile of cases, there are still more tourists than air
passengers (-85 trips). Table 2 also reveals that, as the distribution is quite stable over time, the
divergence between the two sources is no coincidence, but does indeed reflect the structural difference
described above in hypothesis (a).
Table 2. The distribution of the difference between tourists and air passengers
Percentiles 2011 2012 2013 2014 2015 2016
Min -89,300,000 -89,800,000 -89,400,000 -90,200,000 -92,400,000 -93,400,000
1% -3,918,997 -4,064,395 -4,002,791 -4,361,469 -3,865,980 -4,136,718
5% -514,371 -581,089 -661,828 -655,484 -569,920 -643,928
10% -192,821 -212,287 -235,487 -232,265 -183,901 -218,354
25% -22,009 -27,635 -30,651 -28,778 -24,436 -28,451
50% -1,997 -2,536 -2,924 -2,493 -2,189 -2,323
75% -63 -113 -126 -94 -56 -55
90% 1,770 1,220 998 1,371 1,480 4,097
95% 11,824 10,775 8,400 10,081 10,992 28,604
99% 131,253 140,405 109,720 113,494 140,005 257,340
Max 1,137,767 834,788 1,070,940 1,191,830 1,396,962 2,525,211
Obs. 5,359 5,771 5,649 5,653 5,779 5,262
Mean -210,505 -219,209 -232,735 -232,250 -224,670 -243,573
Std. Dev. 2,175,910 2,132,248 2,178,926 2,221,919 2,251,131 2,393,686
Skewness -30 -30 -28 -28 -29 -27
Kurtosis 1,105 1,131 1,048 1,020 1,043 939 Note: Negative values denote that there are more tourists than air passengers; positive values denote that there are
more air passengers than tourists travelling between a pair of countries.
Figure 3 shows the relation between the tourist-air passenger discrepancy and geographic distance
(based on CEPII’s GeoDist dataset [Mayer and Zignago 2006]). A clear pattern emerges: there are
sizeable discrepancies at short geographic distances only. The most extreme negative deviations (i.e., a
lot more tourists than air passengers) are Hong Kong ↔ China (89-93 million, depending on year and
direction), Macao ↔ China (37-43 million), United States ↔ Mexico (30-34 million), and Germany ↔
Poland (26-33 million). As Figure 3 clearly shows, extreme cases consistently cluster together over time
(the rings of different colors represent the different years). This suggests that these discrepancies are not
random but systematic and meaningful. The inspection of specific cases with the highest negative8
deviations helps understand the rationales of the discrepancies, which can overlap and reinforce each
other:
1. Mobility between nearby countries: tourists exceed air passengers because many people move
across borders with land (train, car, bus) or water (ferry, ship) transportation. Examples include
the four extreme outlier country pairs tagged in Figure 3.
2. Grand tour tourism: Here, people fly to one country (e.g., from the U.S. to the Netherlands), and
then go by car or train to other countries (e.g., France). In France, they are counted as tourists
(e.g., through hotel registration data) but not as air passengers.
8 In fact, there are few exceptional cases in which air passengers are in larger numbers than registered tourists. These are
mostly distant countries with large contingents of migrants or returning nationals (who are not registered by tourism
statistics) but relatively modest inflows of other visitors (e.g., India and Oman).
Estimating Transnational Human Mobility on a Global Scale
European University Institute 7
While rationale (2) is difficult to deal with (see the remaining limitations described in section 5), we
treat rationale (1) by creating a correction factor that takes distance into account.
Figure 2. Cumulative distributions of the difference between the GMP-revised tourism dataset
[1] and the KCMD-revised air passenger trend dataset [2]
Note: ECDF = Empirical cumulative distribution function
3.3 Creating the distance-adjusted air passenger data [3]
The goal here is to adjust the KCMD-revised air passenger trend data [2] to correct for the fact that it
underestimates mobility at short distances due to the use of alternative transportation means. To do so,
we draw on the distance (in km) between country pairs. Our correction factor is specified as:
(𝑘𝑚𝑎𝑥
𝑘𝐴↔𝐵)
1𝑐⁄
where 𝑘𝑚𝑎𝑥 is the maximum possible distance between two countries, in this case 19,951.16 km (the
distance between Paraguay and Taiwan), and 𝑘𝐴↔𝐵 is the empirical distance between two countries A
and B, based on CEPII’s GeoDist dataset (Mayer and Zignago 2006). The parameter c is chosen so that
it maximizes the correlation r between the GMP-revised tourism data [1] and the KCMD-revised air
passenger trend data [2].9 The rationale behind this is the assumption that [1] is not biased in terms of
9 We combine data from all available years and exclude cases with more than 10 million trips to reduce the influence of these
outliers on the calculations. On average, 31 cases are ignored per year (0.08 percent of the total).
Ettore Recchi, Emanuel Deutschmann and Michele Vespe
8 Robert Schuman Centre for Advanced Studies Working Papers
distance. Distance-adjusting [2] so that its correlation with [1] is maximized should thus lead to the best
possible correction factor.
Figure 3. The relation between geographic distance and divergences between the GMP-revised
tourism dataset [1] and the KCMD-revised air passenger trend dataset [2]
Note: Different colors denote different years. Distance is obtained from Mayer and Zignago (2006)
Estimating Transnational Human Mobility on a Global Scale
European University Institute 9
Figure 4. Adjusting the distance-based correction factor for the KCMD-revised air passenger
trend data to maximize the fit with the GMP-revised tourism data
Ettore Recchi, Emanuel Deutschmann and Michele Vespe
10 Robert Schuman Centre for Advanced Studies Working Papers
Figure 5. The correlation between the distance-adjusted air passenger data [3] and the GMP-
revised tourism data [1]
3.4 Creating the Global Transnational Mobility Dataset
In the final step, we merge the two revised data sources. As we hold that both the GMP-revised tourism
data [1] and the distance-adjusted air passenger data [3] individually tend to under-estimate actual
mobility flows (see section 2), our final estimate is always the largest of the two when we have both
information—that is, either [1] or [3]. When we lack [3], we take [1]; and vice versa. As final steps, we:
- Round decimals (non-integer estimates can occur due to the time-series decomposition
applied by Gabrielli et al [2019] and the correction factor introduced in section 3.3).
- Add missing full country names and information on the world region a country is situated in
based on the United Nations classification (drawing on Duncalfe [2018]).
- We exclude countries for which, after the merging procedure, no information was available.10
Consequently, the dataset is reduced to the set of 196 countries used when creating the
unified UNWTO dataset.
10 Countries and territories excluded are: Aruba, Anguilla, Cocos Islands, Cook Islands, Christmas Islands, Western Sahara,
Falkland Islands, Faroe Islands, Guadeloupe, Grenada, Greenland, French Guiana, Montenegro, Northern Mariana Islands,
Montserrat, Martinique, New Caledonia, Norfolk Islands, Pitcairn, Puerto Rico, French Polynesia, Reunion, Saint Helena,
Saint Pierre and Michelon, Serbia, Tokelau, Taiwan, Wallis and Futuna Islands.
Estimating Transnational Human Mobility on a Global Scale
European University Institute 11
The resulting Global Transnational Mobility Dataset can be explored on an interactive world map at the
KCMD Dynamic Data Hub (https://bluehub.jrc.ec.europa.eu/migration/app/index.html; browse
‘Datasets’ – ‘Mobility (JRC)’ – ‘Estimated Trips (KCMD-EUI)’). More information on the website of
the Migration Policy Centre of the EUI (http://www.migrationpolicycentre.eu/globalmobilities/). The
dataset can be requested for scientific research by email (GMPdataset@eui.eu). It contains the following
variables:
Table 3. Variables contained in the Global Transnational Mobility Dataset
Name Description
source_name Name of the country of origin
target_name Name of the country of destination
source_iso3 ISO 3166-1 alpha-3 code of the country of origin
target_iso3 ISO 3166-1 alpha-3 code of the country of destination
year Year, ranges from 2011 to 2016
estimated_trips Estimated trips
dist Geographic distance
source_region Region of the country of origin
target_region Region of the country of destination
source_subregion Sub-region of the country of origin
target_subregion Sub-region of the country of destination
Note: Geographic distance is obtained from CEPII’s GeoDist dataset (Mayer and Zignago 2006). Regions and
subregions are based on the UN M.49 GeoScheme.
4. Exploring the dataset: some first insights
The Global Transnational Mobility Dataset covers 196 sender and receiver countries and, through the
integration of two different sources, is more comprehensive than all pre-existing information on
worldwide cross-border mobility. This is illustrated in Figure 6, which displays the estimates of mobility
given by several sources. According to UNHCR, there were 2.8 million new asylum-seekers crossing
borders globally in 2016. The number of yearly migrant flows is very difficult to establish, but according
to one estimate, it could be around 8 million people per year.11 The global stock of refugees is estimated
to be 22.5 million for 2016 (UNHCR 2016). In the original UNWTO tourism files, around 1.3 billion
tourist trips are recorded. A similar number is obtainable from the KCMD-revised air passenger trend
data. According to our new dataset, there were about 2.9 billion cross-border trips in 2016.
11 This figure is based on Abel and Sander (2014) and is obtained by dividing the estimate of global migration flows from the
mid-2005 to mid-2010 period by 5. Estimates for more recent years are unfortunately unavailable.
Ettore Recchi, Emanuel Deutschmann and Michele Vespe
12 Robert Schuman Centre for Advanced Studies Working Papers
Figure 6. Comparison between estimates
Note: Sources: Tourists: UNWTO (2016); migrants: Abel and Sander (2014) (estimate of global migration flows
from the mid-2005 to mid-2010 period divided by 5); refugees and asylum-seekers: UNHCR (2015); air
passengers: KCMD-revised air passenger trend data. Note that the unit differs between sources: asylum-seekers,
migrants and refugees are mobile persons, whereas tourists, air passengers and Global Transnational Mobility
(GTM) data are recorded in cross-border trips.
While we leave to future research the full exploitation of its potential, also in conjunction with other
datasets, a preliminary exploration of the Global Transnational Mobility Dataset offers several major
insights that are detailed in this section.
4.1 Worldwide transnational mobility is rapidly increasing over time
Figure 7 shows that during the time frame under study, 2011 to 2016, transnational human mobility
increased dramatically. In absolute terms, the number of estimated trips increased from about 2.3 billion
in 2011 to about 2.9 billion in 2016 (Figure 7A). As Figure 7B reveals, this growth is much larger than
the growth in world population, indicating that collectively, humanity has indeed become more
transnationally mobile. In this regard, transnational mobility is developing similarly as cross-border
communication, but differently from migration, which has not grown significantly faster than the world
population (Deutschmann 2016).
This development raises questions for many fields of inquiry, like the environmental consequences, the
potential spread of epidemics, the emergence of global systemic risks (Centeno et al. 2015) and, from a
sociological perspective, the social inequalities in access to these increased mobility chances. The latter
issue is briefly touched upon in the following section.
Estimating Transnational Human Mobility on a Global Scale
European University Institute 13
Figure 7. Absolute and relative growth of global mobility
Note: The graphs are based on the Global Transnational Mobility Dataset (trips) and World Bank (2018)
population data.
4.2 Transnational mobility tends to cluster within world regions
Figure 8A shows the mobility (in million trips) within world regions, using the United Nations M.49
Geoscheme as a base for assigning countries to regions. We find that Europe is the region with the
highest number of intraregional trips, followed by Asia. The Americas are behind, and the smallest
number of trips occur within Africa and Oceania. Over time, between 2011 and 2016, intraregional
mobility grows strongly in Europe and Asia. The Americas see a smaller increase and mobility in Africa
and Oceania looks much more stable in comparison. There is thus no clear catch-up effect. Rather the
divergence between regions in terms of intraregional mobility seems to widen over time.12
Interregional mobility can be studied by either taking the outgoing mobility from a specific region
(Figure 8B) or the incoming mobility to a specific region (Figure 8C) into account. Both strategies yield
very similar outcomes. In both cases, interregional mobility is far less common than intraregional
mobility, at least for Europe, Asia, and the Americas (cf. Figure 9 and its discussion below). Also note
that the order between regions is the same in terms of intra- and interregional mobility.
Figure 9 allows us to take a closer look at the ratio of intra- to interregional mobility by region. This
could be described as a measure of relative regionalism (Deutschmann 2017). This indicator reveals a
12 Note that this simple measure may not be the best one to study how regionalized mobility actually is. It is well possible
that within Europe, for example, the high number of trips is driven by a subset of country pairs and that others participate
very little in the intraregional network of transnational human mobility. Deutschmann (2017) proposes to use density-based
measures as an alternative that allows to take into account between how many country pairs in a region meaningful amounts
of mobility exist. Moreover, more sophisticated analyses would have to consider the varying population sizes of regions.
Ettore Recchi, Emanuel Deutschmann and Michele Vespe
14 Robert Schuman Centre for Advanced Studies Working Papers
very similar picture regardless of whether incoming or outgoing mobility is used as a measure. In both
cases, intraregional mobility is more than 5 times more likely to occur than interregional mobility in the
case of Europe, more than 4 times in the case of Asia, and almost 3 times in the case of the Americas.
In the case of Africa intraregional mobility is basically as likely as interregional mobility, and in
Oceania, intraregional mobility is even half as likely as interregional mobility.
Note, however, that this comparison may be seen as ‘unfair’ since the pool of potential connections
is obviously much larger in the case of interregional mobility than in the case of intraregional mobility.
A more sophisticated and ‘just’ comparison (which goes beyond the scope of this paper) would be to
compare intraregional mobility to mobility towards specific other world regions. Past research has found
that when this is done, mobility also tends to cluster within Africa and Oceania (Ibid.).
In any case, Figures 8 and 9 highlight the extreme stratification of the chance to engage in
transnational mobility at the global scale. Transnational mobility within Europe is about 20 times the
amount of mobility within Africa, in spite of the much larger population of the latter continent. This
global inequality in mobility chances has important sociological implications. For example, it has been
shown that transnational human capital is an important resource that increases life chances (Gerhards et
al. 2017). Furthermore, transnational mobility shapes world views, attachment to other countries and
cosmopolitan attitudes (Mau et al. 2008; Helbling and Teney 2015; Kuhn 2015; Recchi 2015;
Deutschmann et al. 2018; Recchi et al. 2019). While these consequences of unequal access to
transnational mobility chances have mainly been studied from a European viewpoint so far, a global
perspective is largely missing. The Global Transnational Mobility Dataset may prove a good starting
point for future analyses in this direction. The next section digs a little deeper into this global
stratification by looking at the relation between transnational human mobility and levels of prosperity.
4.3 Transnational mobility differs by levels of prosperity and country size
Figure 10 illustrates how transnational mobility differs by levels of prosperity and country size. Figure
10A shows a clear relation between a country’s outgoing trips and the national level of prosperity,
measured as GDP per capita in purchasing power parity (World Bank data). The relation is relatively
strong and significant, with a correlation coefficient of r = .63. Figure 10B shows a similar pattern for
the relation between mobility and population size. Again, the correlation is quite high with r = .58. The
three-dimensional graph in Figure 10C illustrates the relation between the three factors in combination.
The distribution of dots, representing countries, follows a clear pattern, ranging from low GDP, small
population and low mobility (blue dots at the bottom front corner) to high GDP, large population and
high mobility (red dots in the upper back corner). These insights are not entirely new (e.g., Deutschmann
2016 and 2017), but are showcased in a clear and robust way by this novel dataset. Future research may
engage in more complex analyses, taking a larger set of factors into account and building more
comprehensive multivariate models to study the antecedents and consequences of transnational human
activity worldwide.
Estimating Transnational Human Mobility on a Global Scale
European University Institute 15
Figure 8. Mobility within and between world regions
Ettore Recchi, Emanuel Deutschmann and Michele Vespe
16 Robert Schuman Centre for Advanced Studies Working Papers
Figure 9. Relative regionalism, by region
Estimating Transnational Human Mobility on a Global Scale
European University Institute 17
Figure 10. The relation between mobility, population size, and GDP per capita.
5. Conclusion
A spate of migration and asylum-seeking crises has been hitting the world since the turn of the 21st
century. The globe is on the move but, in spite of their over-exposure in the media and public opinion,
migrants and refugees constitute only a tiny portion of the whole number of people crossing borders
daily. According to Abel and Sanders’ (2014) estimates, there were less than 10 million worldwide
migration episodes per year in the early 2010s worldwide. According to our estimate, yearly border-
crossings come close to 3 billion globally. By providing estimates of the amount of such transnational
mobility beyond migration, the Global Transnational Mobility Dataset—created as an outcome of this
paper—facilitates the study of the volume, directions and change of country-to-country human mobility
on a worldwide scale.
This paper has described the procedures by which we have reached these estimates. While we
acknowledge that there is no single existing data source providing exact information on the number of
people officially crossing national borders worldwide, we do find that the two more complete and
reliable sources (data on tourism and data on air passengers) do show significant consistency and can
be merged according to a few and relatively simple combination rules.
Focusing on yearly country-to-country flows of human mobility (whatever their duration), our dataset
complements estimates of worldwide migration flows (Abel and Sanders 2014), which refer to stays
abroad longer than 12 months, based on the conventional UN definition of migration. This dataset also
advances previous usages of the UNWTO data (Reyes 2013; Deutschmann 2016 and 2017), capitalizing
on an additional source and estimation methods. Finally, the Global Transnational Mobility Dataset
Ettore Recchi, Emanuel Deutschmann and Michele Vespe
18 Robert Schuman Centre for Advanced Studies Working Papers
parallels recent and alternative attempts at measuring population mobility with digital sources (State et
al. 2013; Hawelka et al. 2014; Messias et al. 2016; Rango and Vespe 2017; Zagheni et al. 2017; Fiorio
et al. 2017; Spyratos et al. 2018). Data triangulation across these digital estimates and ours may prove
useful to test the comparability of outcomes obtained through such different approaches.
Several important limitations remain. The first issue concerns the existence of grand-tour tourism
and open-jaw flights (see section 3.2). For instance, consider a traveler who goes on a round trip to
Southeast Asia from Italy. She flies from Rome to Bangkok both on her way in and out and takes buses
or rents a car to travel subsequently through Thailand, Vietnam, Laos, and Cambodia, before returning
to Thailand to take her flight back home. According to the original UNWTO tourism data, there would
be four trips: ITA→THA, ITA→VNM, ITA→LAO, and ITA→KHM. According to the GMP-revised
tourism data [1], there would be eight trips: ITA→THA, THA→ITA, ITA→VNM, VNM→ITA,
ITA→LAO, LAO→ITA, ITA→ KHM, and KHM→ITA. According to the air passenger data
(regardless of distance-adjustment), there would be two trips: ITA→THA, THA→ITA. In reality,
however, there were six trips: ITA→THA, THA→KHM, KHM→VNM, VNM→LAO, LAO→THA,
and THA→ITA. In this case, both sources and all strategies lead to very different outcomes and none
of them captures the transnational mobility that actually took place. This is an issue that has no easy
solution. Structurally, it should lead to a slight overestimation of long-distance mobility between world
regions (which is most likely when such round-trips are prone to occur). However, we argue that,
compared to all global travels, this kind of journeys are rare and should not jeopardize the overall
reliability of the dataset.
A second limitation is the following: by basing a substantial part of our mobility estimates on visitors
who stayed overnight (‘tourists’ in the UNWTO terminology), we may be underestimating short-term
border-crossings, for instance by commuters who live in border regions and regularly cross to the other
side for work, leisure, or shopping. The following example is revealing in this regard: For the USA,
detailed data on land border crossings are available (US Department of Transportation 2018). Looking
at mobility between the USA and Canada, the distance-adjusted air passenger data [3] estimates about
20 million trips, while the GMP-revised tourism data [1] suggests around 33 million trips. The recorded
land border crossing, by contrast, are 103 million—98 million private car passengers alone. Many of
these moves are likely not overnight stays. While it is hard to generalize from this example,13 it suggests
that the mobility estimates in the Global Transnational Mobility Dataset (and the correction factor
introduced in section 3.3)—although considerably larger than those provided by alternative global
sources—are still quite conservative.
Finally, it is important to keep in mind that what the Global Transnational Mobility Dataset contains
are mobility estimates rather than counts of actual, recorded trips. This is crucial. By applying a
statistical approach to correct and adjust the data, we aimed at creating a revised dataset that on average
captures mobility between countries more accurately. This procedure can however imply that in a
minority of individual cases this revision leads to a more inaccurate estimate. We would thus like to
remind that this dataset is well-suited to study structural features of transnational human mobility
globally or for aggregates of countries. If the research interest is mobility between specific pairs of
countries, the estimates in the Global Transnational Mobility Dataset are to be taken with caution, being
aware of this limitation, and possibly comparing them to figures provided by alternative sources.
With these caveats in mind, we maintain that this novel dataset will prove to be a valuable resource
for researchers interested in studying the global structure of transnational human mobility and its links
to phenomena in the social and natural world, from wealth and well-being to the spread of epidemics
and climate change.
13 Table A1 in the Appendix provides an overview of the small number of cases in the UNWTO data where both ‘visitors’
and ‘tourists’ (i.e., overnight visitors) are reported. ‘Tourists’ as a share of ‘visitors’ range from 2 to 98 percent. The
variance in this regard across countries is thus huge.
Estimating Transnational Human Mobility on a Global Scale
European University Institute 19
Appendix: Further details regarding the GMP revision of the UNWTO files
In section 3.1, we described the revision of the raw UNWTO data. In the following, several additional
details regarding this process are given. First, several decisions had to be made to derive the preference
order that is used when several categories of ‘arrivals’ are available in the same receiver country file (cf.
Table 1 in the main text).
Issue 1: ‘by nationality’ vs ‘by country of residence’
For almost all receiver countries, arrivals are reported either ‘by nationality’ or ‘by country of
residence’. In the few cases where both are reported,14 we found that values do not differ dramatically
between the two categories. If we were to decide for a restriction to one of these categories, we would
lose a large percentage of cases (see category [1] vs [3] or [2] vs [4] in Figure A1). These two aspects
taken together justify merging arrivals reported ‘by nationality’ and ‘by country of residence’ in a single
dataset. In the rare cases where both categories are available, preference was given to ‘by country of
residence’.
Issue 2: ‘tourists’ vs ‘visitors’
For a relative large percentage of cases, data on ‘tourist’ arrivals are unavailable and data on ‘visitors’
is reported instead (Categories [2] and [4] in Figure A1). We believe the benefit of not losing these cases
outweighs the drawback of the imprecision that results from merging the two different categories in one
dataset. According to the UNWTO definition (see section 3.1), ‘visitor’ is a broader category that
includes ‘both tourists and same-day non-resident visitors’. There are very few cases where country-to-
country arrival data on both tourists and visitors is available (Table A1). However, in such cases, the
size of the difference varies largely. In Venezuela, tourists as a share of visitors constituted 98 percent
in 2010, whereas in Belarus it was only 2 percent, with the other thirteen countries being distributed
quite evenly across the whole percentage range in between. (It seems plausible that in small countries
the difference is more sizeable than in large countries). In the rare case that both ‘tourists’ and ‘visitors’
are reported we give preference to ‘tourists’ since the majority of cases are reported as tourists
(categories [1], [3] and [5-8] in Figure A1, making it more or less the ‘standard category’).
Issue 3: ‘at national borders’ vs ‘in accommodation establishments’ A third issue concerns the question of whether data collected ‘at national borders’ is comparable to data
collected via ‘accommodation establishments’ (Table A2). To get an idea, we can draw on a total of 20
receiver countries for which both category types are available. In 17 out of these 20 countries, the
number of arrivals at national borders is larger than the number of arrivals in accommodation
establishments. A likely explanation is that some travelers who arrive in the country find private
accommodation that is not covered in the data. In three exceptional cases (Iceland, Israel, Thailand),
there are more arrivals in accommodation establishments than at national borders. On average, i.e.,
across all 460 cases (i.e., country-years) for which data is available, the ratio is .786, which could be
interpreted as: on average, the number of arrivals reported for accommodation establishments is 78.6
percent the size of the number of arrivals reported at national borders. Note however, that the according
standard deviation is very large (.456 or 46.5 percent) which makes the meaning and usability of this
mean value questionable. The across-time variance within countries is much smaller (.085 or 8.5 percent
on average), suggesting that individual countries are relatively consistent in their reporting style, while
between countries there are considerable differences.
14 These cases are, for 111 vs. 112: Guinea, Mali, Mexico, Nepal, Sri Lanka, and Thailand, for 121 vs 122: Indonesia, Macao,
and Singapore. For 1911 vs. 1912 and 711 vs 712 no countries with both categories reported were found. In the cases of
Guinea, Nepal, Mexico, Indonesia, and Macao, information was more detailed in the 111 and 121 categories than in the
112 and 122 categories.
Ettore Recchi, Emanuel Deutschmann and Michele Vespe
20 Robert Schuman Centre for Advanced Studies Working Papers
Figure A1: Distribution of arrival categories in the 196-country version of the UNWTO dataset
Note: 1= Arrivals of non-resident tourists at national borders, by nationality; 2 = Arrivals of non-resident visitors
at national borders, by nationality; 3 = Arrivals of non-resident tourists at national borders, by country of residence;
4 = Arrivals of non-resident visitors at national borders, by country of residence; 5 = Arrivals of non-resident
tourists in all types of accommodation establishments, by country of residence; 6 = Arrivals of non-resident tourists
in hotels and similar establishments, by nationality; 7 = Arrivals of non-resident tourists in hotels and similar
establishments, by country of residence; 8 = Arrivals of non-resident tourists in all types of accommodation
establishments, by nationality
Table A1. Tourists as a share of all visitors in 15 countries with both categories available, 2010
Country Tourists Visitors Tourists as a share of
visitors
Belarus 118,749 6,129,863 2%
Belize 241,919 1,197,326 20%
Hungary 9,511,000 39,905,000 24%
British Virgin Islands 330,343 842,497 39%
Jordan 4,207,408 8,078,380 52%
Hong Kong 20,085,155 36,030,331 56%
Italy 43,626,118 73,225,219 60%
Canada 16,219,399 25,621,300 63%
South Africa 8,073,552 11,303,087 71%
Israel 2,803,125 3,443,988 81%
Mongolia 456,963 557,452 82%
Namibia 984,098 1,114,423 88%
Saint Vincent and the Grenadines 72,478 77,564 93%
Turkey 31,364,004 32,997,308 95%
Venezuela 526,255 535,270 98% Note: for Belarus, 2012 was used since 2010 was missing
1
2
3
4
8
7
6
5
Estimating Transnational Human Mobility on a Global Scale
European University Institute 21
Table A2. Arrivals ‘in accommodation establishments’ as a share of ‘at national borders’
Categories of comparison/receiver country Mean
SD across
years
1911 as a share of 111 (‘all types of accommodation’, ‘by nationality’)
Hungary 0.376 0.024
Iceland 2.412 0.175
Italy 0.975 0.078
Turkey 0.652 0.140
Cross-country mean 1.104 0.105
1912 as a share of 112 (‘all types of accommodation’, ‘by country of residence’)
Cyprus 0.779 0.050
Philippines 0.985 0.004
Spain 0.716 0.117
France 0.530 0.027
Greece 0.549 0.051
Cross-country mean 0.712 0.050
711 as a share of 111 (‘hotels, etc.’, ‘by nationality’)
Thailand 1.737 0.083
Hungary 0.342 0.024
Iceland 1.658 0.085
Italy 0.785 0.051
Turkey 0.647 0.142
Morocco 0.577 0.185
Tunisia 0.769 0.210
Chad 0.337 0.063
El Salvador 0.481 0.144
Bolivia 0.765 0.108
Cross-country mean 0.810 0.109
712 as a share of 112 (‘hotels, etc.’, ‘by country of residence’)
Guinea 0.390 0.080
Mali 0.293 0.140
Cyprus 0.777 0.051
Philippines 0.374 0.055
Spain 0.587 0.080
France 0.427 0.033
Greece 0.536 0.051
Norway 0.876 0.035
Israel 1.114 0.165
Malta 0.773 0.007
Cross-country mean 0.615 0.070
Mean of all country-means 0.776 0.085
SD across all country-means 0.469
Global Mean/SD across all 460 country-years 0.786 0.465
Note: The underlying data stems from the whole time range, i.e., 1995 to 2016. Figures in red refer to countries in which
exceptionally arrivals recorded in accomodation establishments are larger than those recorded at national borders
Ettore Recchi, Emanuel Deutschmann and Michele Vespe
22 Robert Schuman Centre for Advanced Studies Working Papers
Table A3. Arrivals in ‘hotels and similar establishments’ as a share of arrivals in ‘all kinds of
accommodation establishments’, 1995-2016
Categories of comparison/receiver country Mean SD across years
711 as a share of 1911 (‘by nationality’)
Hungary 0.865 0.066
Iceland 0.694 0.039
Italy 0.811 0.021
Turkey 0.991 0.010
Czech Republic 0.904 0.048
Slovenia 0.776 0.045
Macedonia 0.908 0.036
Cross-country-mean 0.850 0.038
712 as a share of 1912 (‘by country of residence’)
Cyprus 0.997 0.004
Philippines 0.380 0.055
Spain 0.828 0.068
France 0.749 0.030
Greece 0.977 0.005
Norway 0.650 0.022
Bulgaria 0.984 0.006
Croatia 0.453 0.041
Estonia 0.927 0.022
Poland 0.839 0.058
Denmark 0.340 0.161
Lithuania 0.867 0.029
Portugal 0.920 0.018
Romania 0.973 0.020
Sweden 0.623 0.036
Austria 0.732 0.005
Belgium 0.792 0.021
Germany 0.891 0.007
Luxembourg 0.732 0.062
Netherlands 0.777 0.030
Switzerland* 0.885 n.a.
Cross-country-mean 0.777 0.035
Mean/SD across all country-averages 0.795 0.036
SD across all country-means 0.176
Global Mean/SD across all 520 country-years 0.794 0.178
Note: *only available for one year
Estimating Transnational Human Mobility on a Global Scale
European University Institute 23
To get the most comprehensive picture possible, we use both categories but give preference to the
category ‘at national borders’ wherever it is available. For the sake of consistency, we do the same in
the exceptional cases of Iceland, Israel, and Thailand. There are 22 receiver countries for which only
data on arrivals at accommodation establishments is reported (i.e., only [one/some of] the categories
711, 712, 1911, 1912 are available):
Austria, Belgium, Bosnia & Herzegovina, Burkina Faso, Cape Verde, Croatia, Czech Republic,
Denmark, Estonia, Germany, Lithuania, Macedonia, Norway, Luxembourg, Netherlands, Palestine,
Portugal, Senegal, Slovakia, Slovenia, Switzerland, Togo.
In order not to lose these receiver countries, we keep them in the data, assigning preference to the
categories as indicated in Table 1 in the main text. Given the calculations described above, it is possible
that for these 22 countries arrivals are underestimated.
Issue 4: ‘all types of accommodation establishments’ vs ‘hotels and similar establishments’
A fourth issue concerns the difference between ‘all types of accommodation establishments’ vs ‘hotels
and similar establishments’. Here, we can draw on 28 countries for which both category types are
available to get an idea of the extent of the difference. Table A3 shows that, as one would expect, ‘all
kinds of accommodation establishments’ is always the larger category. Across all 520 cases (i.e.,
country-years) for which we have data, arrivals in ‘hotels and similar establishments’ are on average
79.5 percent the size of arrivals in ‘all types of accommodation establishments’. Note, however, that
there is quite some variance between countries, with the share ranging from 34.0 percent in Denmark to
99.7 percent in Cyprus. The standard deviation across all country-years is .178 or 17.8 percent. To get
the most comprehensive picture, we give preference to the category ‘all types of accommodation
establishments’ whenever it is available.
Due to (a) the large variance between countries, which makes the average share rather meaningless
and (b) the fact that most countries from which we could make inferences are European while most
countries for which we lack information are African (which may result in deviating reporting styles),
we refrain from using the information given in Table A.3 to create a factor to correct for the likely
underestimation of the number of arrivals in five countries for which only arrivals in ‘hotels and similar
establishments’ are reported. These countries are Burkina Faso, Cape Verde, Palestine, Senegal, and
Togo. This implies that for these five receiver countries arrivals are likely underestimated.
Issue 5: dealing with ‘odd’ travel origin categories
Besides bringing order into the various ‘arrival’ categories, there are several ‘odd’ categories of origin
of travels in the data that need to be dealt with. Their relative weights in the full dataset are shown in
Table A4.
Table A4. ‘Odd’ categories of travel origin in the UNWTO data
Category Percentage
1. Normal cases (e.g., ‘Albania’) 92.5
2. Country pairs (e.g., ‘Canada, United States’) 2.7
3. ‘Nationals residing abroad’ 1.0
4. ‘USSR (former)’; ‘Scandinavia’; ‘Yugoslavia, SFR (former)’; ‘Benelux’ (6 cases) 0.01
5. ‘Other countries of [world region, ‘the world’]’ 2.9
6. ‘All countries of [world region]’ 0.9
All ‘odd’ categories 7.5
Lost arrivals after measures taken approx. 3.8
Note: Percentage refers to the total number of tourist arrivals, not to the number of cases.
Ettore Recchi, Emanuel Deutschmann and Michele Vespe
24 Robert Schuman Centre for Advanced Studies Working Papers
Category 1. Normal cases
The vast majority of cases (92.5 percent) are ‘normal’ cases, i.e., they state the number of arrivals from
a specific sender country to a specific receiver country. They are thus in the appropriate format to be
considered in a country-to-country flow matrix.
Category 2: Country pairs
In 45 cases, the sender is not an individual country, but one of seven country pairs:
‘Australia, New Zealand’; ‘Belgium / Luxembourg’; ‘Canada, United States’; ‘China + Hong Kong,
China’; ‘Czech Republic/Slovakia’; ‘Serbia and Montenegro’; ‘United Kingdom/Ireland’.
In order not to lose these cases (which include major sender country pairs such as ‘Canada/United
States’, we split the number of arrivals reported for these cases into portions corresponding to the
population size of the two sender countries in the according year weighted by the two countries’
populations’ general propensity to get involved in tourism. This general propensity to get involved in
tourism is calculated from the overall number of arrivals from that country in all normal cases (i.e.,
Category 1).
Category 3: Nationals residing abroad
For 29 receiver countries, the sender category ‘nationals residing abroad’ is reported. These countries
include:
Algeria, Belize, Burkina Faso, Chile, Colombia, Congo DR, Cuba, Dominican Republic, Gambia,
Grenada, Guinea, Iran, Jordan, South Korea, Mexico, Morocco, New Zealand, Nicaragua, Nigeria,
Oman, Philippines, Rwanda, San Marino, Saudi Arabia, Togo, Tunisia, Turkey, Uruguay, and
Yemen.
Since no clear country of origin (of the trip) can be identified for these cases, we decided to drop them.
Categories 4-6: Broad group of countries
Regarding categories 4-6, there are two main obstacles. First, the assumption that the tourists will be
split according to the population distribution and their propensity to engage in tourism becomes rather
questionable for such large groups of countries (think of ‘other countries of the world’), and hard to
compute. Furthermore, it would require determining, for each case in category 5-6, which countries were
not listed from a specific world region since this varies from receiver country to receiver country. These
efforts combined with the questionable quality of the outcome seem to justify neglecting these categories
rather than imposing problematic assumptions about them. Accordingly, we drop and ignore these cases.
Following all the above-mentioned steps, the number of ‘lost’ arrivals (i.e., not imputable to any
sending country) is reduced to 3.8 percent of all arrivals in the full original version of the dataset. It is
important to note that these 3.8 percent of arrivals are likely to be not randomly distributed. Instead,
most of them result from residual categories (e.g., ‘Other countries in the world’). These residual
categories are presumably often constructed when there are relatively few incoming visitors from distant
parts the world. Thus, we assume that the lost cases are overwhelmingly long-distance travel.
To increase the comparability with the air passenger dataset, we excluded the following countries
and territories:
American Samoa, Anguilla, Aruba, Bonaire, British Indian Ocean Territory, Channel Islands,
Christmas Island, Cocos (Keeling) Islands, Cook Islands, Curaçao, Democratic Yemen (former),
Faeroe Islands, Falkland Islands (Malvinas), French Guiana, French Polynesia, Greenland, Grenada,
Guadeloupe, Guam, Hawaii, Holy See, Isle of Man, Johnston Island, Liechtenstein, Martinique,
Midway Islands, Montserrat, Netherlands Antilles, New Caledonia, Norfolk Island, Northern
Mariana Islands, Pitcairn, Puerto Rico, Reunion, Saba, Saint Helena, Saint Pierre and Miquelon,
Serbia, Sint Eustatius, Sint Maarten (Dutch part), South Sudan, Svalbard and Jan Mayen Islands,
Estimating Transnational Human Mobility on a Global Scale
European University Institute 25
Taiwan, Tokelau, United States Virgin Islands, Wake Island, Wallis and Futuna Island, Western
Sahara.
What remains is a comprehensive set of 196 sender and receiver countries that also underlies the data
used in Deutschmann (2016 and 2017).
Finally, as an overview for researchers interested in exploring the UNWTO tourism files more
closely, we report the availability of categories of arrival by receiver country in Table A5.
Table A5. Categories of arrivals in the UNWTO dataset by receiving country
Co
un
try
11
1
12
1
11
2
12
2
19
12
71
1
71
2
19
11
10
11
10
12
21
11
21
12
No
te
Afghanistan
Albania
X
Algeria
X
Andorra
X
Angola
X
Antigua and
Barbuda
X
Argentina X
Armenia
X
Australia
X
Austria
X
X
X
X
Azerbaijan
X
X
X
Bahamas
X
X
Bahrain
X
Bangladesh X
Barbados
X
Belarus X X
121 only since 2012
Belgium
X
X
X
X
Belize X X
111 only since 1998
Benin
X
Bermuda
X
X
Bhutan X
X
Bolivia X
X
X
111 only since 2006,
711 complete
Bosnia &
Herzegovina
X
X
Botswana
X
Brazil
X
British Virgin
Islands
X X
Brunei
Darussalam
X
Bulgaria
X X
X
X
X
Burkina Faso
X
X
Burundi X
Cambodia
X
Cameroon
X
X
X
Canada
X X
X
Cape Verde
X
X
Ettore Recchi, Emanuel Deutschmann and Michele Vespe
26 Robert Schuman Centre for Advanced Studies Working Papers
Cayman
Islands
X
Central
African
Republic
X
Chad X
X
X
Chile X
China
X
Colombia
X
121 empty
Comoros X
Congo DR X
Congo R
X
X
X
Costa Rica X
Croatia
X
X
X
X
Cuba
X
Cyprus
X
X
X
X
X
Czech
Republic
X
X X
X
Denmark
X
X
X
X
Djibouti
Dominica
X
Dominican
Republic
X
Ecuador
X
Egypt
X
X
El Salvador X
X
X
X
Equatorial
Guinea
Eritrea
X
Estonia
X
X
X
X
Ethiopia
X
Fiji
X
X
Finland
X X
X
X
France
X
X
X
X
X
Gabon X
Gambia X
Georgia
X
X
Germany
X
X
X
X
Ghana X
Gibraltar
no file
Greece
X
X
X
X
Guatemala
X
121 empty
Guinea X
X
X
Guinea-Bissau X
Guyana
X
Haiti
X
Honduras X
Hongkong
X X
112 only since 1998
Hungary X X
X
X X
X
Iceland X
X
X X
X
India X
Indonesia
X
X
X
Iran
X
Iraq
X
Ireland
X
X
Israel
X X
X
X
Italy X X
X
X X
X
Estimating Transnational Human Mobility on a Global Scale
European University Institute 27
Ivory Coast
X
112 empty
Jamaica
X
X
Japan
X
Jordan X X
Kazakhstan
X
Kenya
X
X
Kiribati X
Kuwait
X
Kyrgyzstan
X
112 empty
Laos
X
Latvia
X X
X
Lebanon X
Lesotho
X
Liberia
Libya
X
Lithuania
X
X
X
X
Luxembourg
X
X
X
X
Macao
X
X
X
X
Macedonia
X
X X
X
Madagascar X
X
Malawi
X
X
Malaysia X
X
112 empty
Maldives X
Mali X
X
X
X
711 , 1011 empty
Malta
X
X
X 112 empty
Marshall
Islands
X
X 111 empty
Mauritania
Mauritius
X
X
Mexico X
X
Micronesia
X
Moldova
X
X
X
Mongolia X X
Morocco X
X
X
Mozambique
X
Myanmar X
X
Namibia X X
Nauru
Nepal X
X
Netherlands
X
X
X
X
New Zealand
X
Nicaragua X
Niger X
Nigeria
X
Niue
X
North Korea
Norway
X
X
X
X 112 empty
Oman
X
Pakistan X
Palau X
X
Palestinian
X
X
Panama
X
Papua New
Guinea
X
112 empty
Paraguay X
Peru
X
X
X
Ettore Recchi, Emanuel Deutschmann and Michele Vespe
28 Robert Schuman Centre for Advanced Studies Working Papers
Philippines
X
X
712 empty
Poland
X
X
X
X
X
Portugal
X
X
X
X 111,121,112,122
empty
Qatar
X
Romania
X X
X
X
X
Russia
X
Rwanda
X
Saint Kitts and
Nevis
X
Saint Lucia
X
Saint Vincent
and the
Grenadines
X X
Samoa
X
San Marino
X
Sao Tome and
Principe
X
Saudi Arabia X
Senegal
X
X
Seychelles
X
Sierra Leone
X
X
Singapore
X
X
Slovakia
X
X
Slovenia
X
X X
X
Solomon
Islands
X
Somalia
South Africa
X X
South Korea
X
Spain
X
X
X
X
X
Sri Lanka X
X
X
Sudan X
Suriname
X
Swaziland
X
X
Sweden
X X
X
X
X 112 empty, 122 only
since 2011
Switzerland
X
X
X
X
Syria
several empty
categories
Tajikistan
X
Thailand X
X
X
X
TimorLeste
X
Togo
X
X
Tonga
X
Trinidad and
Tobago
X
112 empty
Tunisia X
X
X
Turkey X X
X
X X
X
Turkmenistan X
Turks and
Caicos Islands
X
Tuvalu X
Uganda
X
Ukraine
X
United Arab
Emirates
Only empty
categories
Estimating Transnational Human Mobility on a Global Scale
European University Institute 29
United
Kingdom
X
X
United
Republic of
Tanzania
X
United States
of America
X
Uruguay
X
Uzbekistan
X
Vanuatu
X
Venezuela X X
Vietnam
X
Yemen X
X
Zambia
X
Zimbabwe
X
Note: To decode the arrival category codes, cf. Table 1 in the main text.
Ettore Recchi, Emanuel Deutschmann and Michele Vespe
30 Robert Schuman Centre for Advanced Studies Working Papers
Author contacts:
Ettore Recchi (corresponding author)
Migration Policy Centre
Robert Schuman Centre for Advanced Studies, EUI
Villa Malafrasca
Via Boccaccio 151
I-50133 Firenze (FI)
and
Sciences Po, Observatoire sociologique du changement (OSC), CNRS, Paris, France
Email: ettore.recchi@eui.eu
Emanuel Deutschmann
Migration Policy Centre
Robert Schuman Centre for Advanced Studies, EUI
Villa Malafrasca
Via Boccaccio 151
I-50133 Firenze (FI)
and
University of Göttingen, Institute of Sociology, Göttingen, Germany
Michele Vespe
European Commission
Joint Research Centre (JRC)
Via Enrico Fermi 2749
21027 Ispra (VA)
Italy
Estimating Transnational Human Mobility on a Global Scale
European University Institute 31
References
Abel, G.J., and N. Sander. 2014. Quantifying Global International Migration Flows. Science 343(6178):
1520-1522.
Centeno, M.A., M. Nag, T.S. Patterson, A. Shaver, and A.J. Windawi. 2015. The Emergence of Global
Systemic Risk. Annual Review of Sociology 41: 65–85.
Deutschmann, E. 2016. The Spatial Structure of Transnational Human Activity. Social Science Research
59: 120–36.
Deutschmann, E. 2017. Mapping the Transnational World: Towards a Comparative Sociology of
Regional Integration. PhD Dissertation, Jacobs University/University of Bremen.
Deutschmann, E., J. Delhey, M. Verbalyte, and A. Aplowski. 2018. The Power of Contact: Europe as a
Network of Transnational Attachment. European Journal of Political Research 57(4): 963–88.
Duncalfe, L. 2018. ISO-3166 Country and Dependent Territories Lists with UN Regional Codes.
Available at: https://github.com/lukes/ISO-3166-Countries-with-Regional-Codes (last accessed
08/01/2019).
Fiorio, L., G. Abel, J. Cai, E. Zagheni, I. Weber, and G. Vinué. 2017. Using Twitter Data to Estimate
the Relationship between Short-term Mobility and Long-term Migration. Proceedings of the 2017
ACM on Web Science Conference, 103-110.
Gabrielli, L., E. Deutschmann, F. Natale, E. Recchi & M. Vespe. 2019. Dissecting Global Air Traffic
Data to Discern Different Types and Trends of Transnational Human Mobility. Unpublished
manuscript.
Gerhards, J., S. Hans, and S. Carlson. 2017. Social Class and Transnational Human Capital. How
Middle and Upper Class Parents Prepare Their Children for Globalization. London/New York:
Routledge.
Hawelka, B., I. Sitko, E. Beinat, S. Sobolevsky, P. Kazakopoulos, and C. Ratti. 2014. Geo-located
Twitter as Proxy for Global Mobility Patterns. Cartography and Geographic Information Science
41(3): 260-271.
Helbling, M., and Teney, C. 2015. The Cosmopolitan Elite in Germany. Transnationalism and
Postmaterialism. Global Networks 15(4): 446–468.
Kuhn, T. 2015. Experiencing European Integration: Transnational Lives and European Identity.
Oxford: Oxford University Press.
Mau, S., J. Mewes, and A. Zimmermann. 2008. Cosmopolitan Attitudes through Transnational Social
Practices? Global Networks 8(1): 1–24.
Mayer, T. and Zignago, S. 2006. GeoDist: The CEPII’s Distances and Geo-graphical Database. MPRA
Paper No. 31243.
Messias, J., F. Benevenuto, I. Weber, and E. Zagheni. 2016. From Migration Corridors to Clusters: The
Value of Google+ Data for Migration Studies. Proceedings of the 2016 IEEE/ACM International
Conference on Advances in Social Networks Analysis and Mining, 421-428.
Nye, J.S., Keohane, R.O. 1971. Transnational Relations and World Politics: An Introduction.
International Organization 25(3): 329-349.
Rango, M., and M. Vespe. 2017. Big Data and Alternative Data Sources on Migration: From Case-
studies to Policy Support. Summary Report. Ispra: Joint Research Centre of the European
Commission.
Ettore Recchi, Emanuel Deutschmann and Michele Vespe
32 Robert Schuman Centre for Advanced Studies Working Papers
Recchi, E. 2015. Mobile Europe: The Theory and Practice of Free Movement in the EU. Basingstoke:
Palgrave Macmillan.
Recchi, E. 2017. Towards a Global Mobilities Database: Rationale and Challenges. Explanatory Note.
MPC/EUI. Available at: http://www.migrationpolicycentre.eu
/docs/GMP/Global_Mobilities_Project_Explanatory_Note.pdf (last accessed 3/3/2019).
Recchi, E., A. Favell, F. Apaydin, R. Barbulescu, M. Braun, I. Ciornei, N. Cunningham, J. Diez
Medrano, D. N. Duru, L. Hanquinet, S. Pötzschke, D. Reimer, J. Salamonska, M. Savage, J. Solgaard
Jensen, A. Varela. 2019. Everyday Europe: Social Transnationalism in an Unsettled Continent.
Bristol: Policy Press.
Reyes, V. 2013. The Structure of Globalized Travel: A Relational Country-Pair Analysis. International
Journal of Comparative Sociology 54(2): 144–70.
Sabre (2014). Aviation Data Intelligence, Leg Flow Tables (2014). Available at:
http://www.sabreairlinesolutions.com/home/software_solutions/airports/ (last accessed 10/3/2019).
Spyratos, S., M. Vespe, F. Natale, I. Weber, E. Zagheni, and M. Rango. 2018. Migration data Using
Social Media: A European Perspective. JRC Technical Report. doi: 10.2760/964282.
State, B., I. Weber, and E. Zagheni. 2013. Studying Inter-national Mobility through IP Geolocation.
Proceedings of the Sixth ACM International Conference on Web Search and Data Mining, 265-274.
United Nations High Commissioner for Refugees (UNHCR). 2016. Global Trends – Forced
Displacement in 2016. Available at: https://www.unhcr.org/globaltrends2016/ (last accessed
9/1/2019).
United Nations World Tourism Organization (UNWTO). 2015. Methodological Notes to the Tourism
Statistics Database. Madrid: UNWTO.
U.S. Department of Transportation. 2018. Border Crossing Entry Data. Available at:
https://data.transportation.gov/Research-and-Statistics/Border-Crossing-Entry-Data/keg4-3bc2 (last
accessed 9/1/2019).
Wimmer, A. and N. Glick Schiller. 2002. Methodological Nationalism and Beyond: Nation-state
Building, Migration and the Social Sciences. Global Networks 2(4): 301-334.
World Bank. 2018. Population, Total. Available at: https://data.worldbank.org/indicator/SP.POP.TOTL
(last accessed 9/1/2019).
Zagheni, E., I. Weber, and K. Gummadi. 2017. Leveraging Facebook's Advertising Platform to Monitor
Stocks of Migrants. Population and Development Review 43(4): 721-734.
Recommended